LETTERdoi:10.1038/nature14001

Predicting immunogenic tumour mutations bycombining mass spectrometry and exome sequencingMahesh Yadav1*, Suchit Jhunjhunwala1*, Qui T. Phung1, Patrick Lupardus1, Joshua Tanguay1, Stephanie Bumbaca1,Christian Franci1, Tommy K. Cheung1, Jens Fritsche2, Toni Weinschenk2, Zora Modrusan1, Ira Mellman1, Jennie R. Lill11

& Lelia Delamarre11

Human tumours typically harbour a remarkable number of somaticmutations1. If presented on major histocompatibility complex classI molecules (MHCI), peptides containing these mutations could po-tentially be immunogenic as they should be recognized as ‘non-self’neo-antigens by the adaptive immune system. Recent work has con-firmed that mutant peptides can serve as T-cell epitopes2–9. However,few mutant epitopes have been described because their discoveryrequired the laborious screening of patient tumour-infiltrating lym-phocytes for their ability to recognize antigen libraries constructedfollowing tumour exome sequencing. We sought to simplify the dis-covery of immunogenic mutant peptides by characterizing their gen-eral properties. We developed an approach that combines whole-exomeand transcriptome sequencing analysis with mass spectrometry toidentify neo-epitopes in two widely used murine tumour models. Ofthe .1,300 amino acid changes identified, 13% were predicted tobind MHCI, a small fraction of which were confirmed by mass spec-trometry. The peptides were then structurally modelled bound to MHCI.Mutations that were solvent-exposed and therefore accessible to T-cellantigen receptors were predicted to be immunogenic. Vaccination ofmice confirmed the approach, with each predicted immunogenic pep-tide yielding therapeutically active T-cell responses. The predictionsalso enabled the generation of peptide–MHCI dextramers that couldbe used to monitor the kinetics and distribution of the anti-tumourT-cell response before and after vaccination. These findings indicatethat a suitable prediction algorithm may provide an approach for thepharmacodynamic monitoring of T-cell responses as well as for thedevelopment of personalized vaccines in cancer patients.

Although CD8 T cells can recognize tumour cells and mediate tumourregression following immunotherapy10, the antigens driving effectiveanti-tumour CD8 T-cell responses remain largely unknown. Tumourantigens can be classified into two categories: tumour-associated self-antigens (for example, cancer-testis antigens, differentiation antigens)and antigens derived from tumour-specific mutant proteins. Since thepresentation of self-antigens in the thymus may result in the eliminationof high-avidity T cells, mutant neo-antigens seem likely to be more im-munogenic. However, identifying these antigens has proved problem-atic, having evaded identification by mass spectrometry, which typicallyrelies on sequence elucidation using public proteomic databases that donot contain patient-specific mutations. Conversely, relying on transcript-omic or exome-sequence analysis for mutation identification followed byMHCI binding prediction algorithms typically yields too many candi-date mutant peptides to be easily evaluated. The strength of our approachlies in combining these two powerful and well-established analytical toolsto identify tumour-associated mutated peptides that are presented onMHCI (Fig. 1a).

We performed whole-exome sequencing on MC-38 and TRAMP-C1mouse tumour cell lines to identify tumour-specific point mutations.Coding variants were called relative to the reference mouse genome to

identify 4,285 and 949 non-synonymous variants in MC-38 and TRAMP-C1, respectively (Fig. 1b). To select for high-confidence mutations andfocus on the mutations likely to be expressed in the majority of the tu-mour cells, we subsequently selected RNA-seq-based variants that werepresent at a minimum of 20% allelic frequency and overlapped with theexome-based variants. This resulted in 1,290 and 67 expressed mutationsin MC-38 and TRAMP-C1, respectively. Next we identified 170 predictedneo-epitopes in MC-38 and 6 predicted neo-epitopes in TRAMP-C1using the NETMHC-3.4 algorithm11 (Fig. 1b, Supplementary Tables 1and 2). Despite high density exome reads, only a small number of non-synonymous variants and predicted neo-epitopes in TRAMP-C1 wereidentified. This low mutational frequency may at least partially explainthe low immunogenicity of TRAMP-C1 observed in vivo (that is, pau-city of tumour-infiltrating lymphocytes (TILs)), compared to MC-38tumours (Extended Data Fig. 1a).

Next, we conducted mass spectrometric analysis for MHCI-presentedpeptides and searched against the transcriptome-generated FASTA data-base (Fig. 1a). This revealed 2,332 unique H-2Kb epitopes and 3,907unique H-2Db epitopes presented in MC-38, and 1,651 unique H-2Kb

epitopes and 1,980 unique H-2Db epitopes presented in TRAMP-C1 cells(Supplementary Table 3). The reduced number of epitopes identified forTRAMP-C1 compared to MC-38 probably reflects weaker MHCI sur-face expression on TRAMP-C1 cells12 (Extended Data Fig. 1b). Overallwe observed that peptides derived from abundant transcripts are morelikely to be presented by MHCI (Fig. 1c, Supplementary Tables 3 and 4),as observed by others13,14.

Of the 1,290 and 67 amino acid changes in MC-38 and TRAMP-C1,respectively, only 7 (7 in MC-38 and 0 in TRAMP-C1) were found to bepresented on MHCI by mass spectrometry (Table 1) after manual val-idation and comparison with a synthetically generated version of thepeptide for spectral accuracy (Extended Data Fig. 2). All but one of theseneo-epitopes were predicted to bind MHCI (half-maximum inhibitoryconcentration (IC50) , 500 nM, Table 1). Both wild-type and mutanttranscripts corresponding to the peptides were expressed by MC-38cells (Supplementary Table 5 and Extended Data Fig. 3) and althoughmost of the exact counterpart wild-type peptides were also predicted tobind MHCI, only three were detected by mass spectrometry (ExtendedData Fig. 4). The mutations were not coded in the germline as thesequence for all 7 genes was confirmed as wild type at the position ofinterest in the genome of a C57BL/6 mouse. These proteins have no knownrole in oncogenesis except Med12, which is frequently mutated in pro-state cancer and in smooth muscle tumours15,16.

The small number of mutated peptides identified by mass spectro-metry compared to the number of peptides predicted to be presentedmay be attributed in part to the sensitivity of the peptide purification andmass spectrometric approach, but also suggests that a limiting factor topresentation could be the peptide generation and transport into theendoplasmic reticulum17. On the other hand, mass spectrometry allows

*These authors contributed equally to this work.1These authors jointly supervised this work.

1Genentech, South San Francisco, California 94080, USA. 2Immatics Biotechnologies GmbH, 72076 Tubingen, Germany.

5 7 2 | N A T U R E | V O L 5 1 5 | 2 7 N O V E M B E R 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

a stringent filter to be incorporated into the workflow by selecting forpeptides with sufficient expression and presentation by MHCI, there-fore ensuring only the peptides most likely to yield an immunologicalresponse are further investigated.

We next asked if immunogenicity of the identified neo-epitopes couldbe predicted. Although there is a correlation between peptide bindingaffinity for MHCI and immunogenicity, other factors also contribute18.For example, interaction of the mutated amino acid with the T-cell an-tigen receptor (TCR) is likely to be essential for the recognition of themutated peptide as ‘foreign’. This is especially true when the counter-part wild-type peptide is also presented on MHCI. Five out of the sevenneo-epitopes exhibited high binding affinity prediction (Table 1, IC50 ,

50 nM). The other neo-epitopes exhibited lower binding affinity pre-diction, suggesting that they might be less immunogenic. We next usedpublished crystal structures of H-2Db and H-2Kb and a Rosetta-basedalgorithm19 to model each of the mutant peptides in complex with MHCIand analyse the potential for the mutant residue in each neo-epitope tointeract with the TCR. In general, TCR recognition of displayed pep-tides is mediated by interactions with peptide residues 3 through 7 (ref. 20).Among the peptides with high binding scores only in the Reps1 andAdpgk peptides did the mutation lie within this range. Structure mod-elling also predicted that the mutated residues were oriented towardsthe solvent interface, and were thus judged to have good potential to beimmunogenic (Table 1 and Fig. 2)20. On the other hand, the mutationsin the Irgq, Aatf and Dpagt1 neo-epitopes were found near the carboxy-terminal end of the peptide, which probably falls outside of the TCRbinding region and suggests that these neo-epitopes were unlikely tobe immunogenic despite the increase in predicted binding affinity toMHCI in comparison to the counterpart wild-type peptide (Table 1and Fig. 2).

We next evaluated the immunogenicity of mutated tumour antigensin vivo by immunizing C57BL/6 mice with peptides encoding the mu-tated epitopes in combination with adjuvant and measured CD8 T-cellresponses using peptide–MHCI dextramers. Three out of six peptideselicited CD8 T-cell responses (Fig. 3a). We had predicted Reps1 and Adpgkto be immunogenic on the basis of structure and binding affinity pre-diction, and both elicited strong CD8 T-cell responses. Of the four pep-tides predicted to be non-immunogenic, only Dpagt1 induced a weakCD8 T-cell response. The exact counterpart wild-type peptides of Reps1and Adpgk were not immunogenic (Extended Data Fig. 5a–d).

We confirmed the immunogenicity of these mutated peptides in thecontext of the MC-38 tumours grown in C57BL/6 mice by analysingTILs and staining with peptide–MHCI dextramers. We found that CD8T cells specific for Reps1, Adpgk and Dpagt1 were enriched in the tu-mour bed (Fig. 3b) but not T cells specific for the other mutant peptides(data not shown). Although there was heterogeneity, Adpgk-specificCD8 T cells were the most abundant of the three, and this was specificto MC-38 tumours as no Adpgk-specific CD8 T cells were detected inTRAMP-C1 tumours (Extended Data Fig. 5e). These results show thatCD8 T-cell responses are generated against neo-epitopes in MC-38 tu-mours and further supports our hypothesis that structural correlationpredicting solvent exposure of the mutation combined with MHCI-binding affinity may provide suitable criteria for triaging potential

Exome coding variations

Predicted neo-epitopes

Spectra-identified neo-epitopes

Predicted immunogenic neo-epitopes

Validated immunogenic neo-epitopes

Transcript coding variations

Exome variations

TRAMP-C1MC-38

Eluted peptides

LC-MS/MS

Immunoprecipitation

(H-2Kb, H-2Db)

Tumour cells

Mutatedpeptides

Immunogenicity

in vivoTranslated

FASTA dB

Prediction ofimmunogenicity

Variant

identificationWhole-exome and

RNA sequencing

a

b

c

0 0,1 1,10 10,100 100+

0

2,500

5,000

7,500

10,000

Num

ber

of

genes

(2)

(5)

0 0,1 1,10 10,100 100+

RPKM interval

0

2,500

5,000

7,500

10,000

Num

ber

of

genes

MHC+MHC+

28,439

4,285

1,290

170

7

2

3

10,118

67

949

6

0

0

0

TRAMP-C1MC-38

MHC–

RPKM interval

MHC–

Figure 1 | Identification of MHCI-presented tumour-specific mutations inMC-38 and TRAMP-C1 tumour cell lines. a, Schematic of the approach foridentifying mutated MHCI-presented peptides using sequencing incombination with mass spectrometry. After immunoprecipitation peptideswere analysed using tandem mass spectrometry and were searched against acustomized FASTA database based on RNA-seq. Immunogenicity ofmutated peptides was further predicted in silico and validated using in vivoimmunogenicity studies. b, Flowcharts for the number of genomic variations orvariant peptides identified at each stage of analysis, and finally the number ofpeptides validated as immunogenic. c, RNA-seq expression profiles of genescorresponding to MHCI-presented peptides. The x axis represents expressionlevels (categorized into five reads per kilobase per million reads mapped(RPKM) intervals). The numbers of genes for each RPKM category thatmanifest MHCI-presented peptide(s) are shown (MHC1), as well as the restof the genes (MHC2). The numbers of mutated MHCI-presented peptidesidentified by mass spectrometry in each RPKM interval are indicatedabove the bars.

Table 1 | Summary of mutant peptides presented by MHCI in the MC-38 cell lineGene Peptide MHC allele IC50 (mut) IC50 (WT) Mutation position Immunogenicity prediction

Dpagt1 SIIVFNL[V/L] H-2Kb 8 34 Anchor (P8) 2Reps1 AQL[P/A]NDVVL H-2Db 9 100 Solvent (P4) 1

Adpgk ASMTN[R/M]ELM H-2Db 2 3 Solvent (P6) 1

Cpne1 SSP[D/Y]SLHYL H-2Db 211 685 Solvent (P4) 2Irgq AALLNSA[G/V]L H-2Db 3 52 Solvent (P8) 2Aatf MAPIDHT[A/T]M H-2Db 30 102 Solvent (P8) 2Med12 DPSSSVLFE[D/Y] H-2Kb 38,300 39,411 No structure 2

The IC50 values (IC50 predicted by NetMHC-3.4 in nM units) are shown. Peptides with IC50 , 500 nM are predicted to bind MHCI. The mutant amino acid’s position is described (whether it was anchored to a MHCmolecule or solvent-exposed), based on homology to known MHC–peptide complex structures. The immunogenicity prediction is simply based on IC50 , 50 nM for the peptide, and position (P3–7: which are likelyto interact with TCR) and solvent exposure of the mutation. Mutation is indicated in bold next to the wild-type amino acid in the bracket in the peptide sequences.

LETTER RESEARCH

2 7 N O V E M B E R 2 0 1 4 | V O L 5 1 5 | N A T U R E | 5 7 3

Macmillan Publishers Limited. All rights reserved©2014

immunogenic peptides. Further analysis of a larger peptide data set willgive a better indication of the statistical utility of this method.

Bulk TILs are usually analysed to monitor frequency and the pheno-type of anti-tumour CD8 T-cell responses, which may not provide a trueassessment because only a fraction of CD8 TILs are tumour-specific6.Using peptide–MHCI dextramers for the three immunogenic peptides,we were able to compare the anti-tumour TILs with the bulk TILs inMC-38 tumours. Although the frequency of tumour-specific CD8 T cellsinfiltrating tumours appear to increase initially, we could not discern aclear correlation between infiltration and tumour growth (Extended DataFig. 6). Interestingly, a vast majority (76.9 6 7.1% s.e.m.) of tumour-specific CD8 TILs, detected using Adpgk–H-2Db dextramer, co-expressedPD-1 and TIM-3, compared to bulk CD8 TILs (52.6 6 3.6%) (Fig. 3c).PD-1 and TIM-3 are inhibitory receptors expressed upon T cell activa-tion and their coexpression on CD8 T cells is associated with chronicantigen stimulation leading to T cell exhaustion21–23. We also observeda population of CD8 TILs (Adpgk–H-2Db dextramer-negative) that ex-pressed PD-1 and TIM-3. This could be due to the effect of the tumourmicroenvironment on CD8 T cells irrespective of their specificity24–26.However, the expression of PD-1 on bulk CD8 TILs was lower thantumour Adpgk-specific CD8 T cells (PD-1 MFI (median fluorescenceintensity), Fig. 3c). These results highlight the differential phenotypiccharacteristics of tumour-reactive cells in the tumour microenvironmentand suggest that tumour-specific CD8 T cells exhibit a more exhaustedstate, in line with recent observations in human tumours6.

To determine if CD8 T cells induced against neo-epitopes could pro-vide protective anti-tumour immunity, healthy mice were immunizedwith the mutated peptides (Adpgk, Reps1 and Dpagt1) and subsequentlychallenged with MC-38 tumour cells (Fig. 4a). Tumour growth was com-pletely inhibited in most of the animals in the vaccine group compared toadjuvant alone (Fig. 4b). Two of the animals that had substantial tumourgrowth in the vaccine group also had the lowest frequency of Adpgk-reactive CD8 T cells in blood before tumour inoculation (Fig. 4c andExtended Data Fig. 7), strongly supporting that CD8 T-cell responsesspecific to mutated peptides conferred protection.

Next, we asked if the neo-epitope-specific CD8 T-cell responses couldbe further amplified in tumour-bearing mice upon immunization (Fig. 4d).

After a single immunization, the frequency of Adpgk-reactive CD8 Tcells increased remarkably in the spleen of tumour-bearing mice com-pared to immunized non-tumour-bearing animals (Fig. 4e). Interestingly,we did not observe a similar enhancement of the CD8 T-cell responseagainst Reps1 and Dpagt1 (Extended Data Fig. 8), suggesting that Adpgkis the immuno-dominant neo-epitope in tumour-bearing mice. We alsoobserved a nearly threefold increase in accumulation of Adpgk-reactiveCD8 T cells among total CD8 TILs in the tumours (Fig. 4e). Peptidevaccination also increased overall infiltration of CD451 cells and CD81

T cells in tumours, which resulted into nearly 20-fold increase in thefrequency of neo-epitope-specific CD8 T cells among the total live cellsin the tumour (Fig. 4f and Extended Data Fig. 9a).

We next analysed the phenotype of peptide-specific CD8 T cells in-duced by peptide vaccination. We found that the frequency of PD-1 andTIM-3 expressing Adpgk-specific CD8 TILs was reduced after vaccina-tion, and the surface expression of PD-1 and TIM-3 on these cells wasalso reduced (Fig. 4g, h). This might be an adjuvant effect as it was alsoseen in the adjuvant alone group. These results suggest that tumour-specific T cells exhibit a less exhausted phenotype after vaccination, andthis was further confirmed by the higher percentage of IFN-c-expressingCD8 and CD4 TILs in the tumours from vaccinated mice (Fig. 4i andExtended Data Fig. 9b).

Protein

MHC haplotype

Mutant

Wild type

Solvent-exposed or

anchor residue

Reps1 Adpgk Dpagt1

H-2Db H-2Db H-2Kb

Solvent Solvent Anchor

AQLANDVVLAQLPNDVVL

ASMTNMELMASMTNRELM

SIIVFNLLSIIVFNLV

Protein

MHC haplotype

Mutant

Wild type

Solvent-exposed or

anchor residue

Cpne1 Irgq Aatf

H-2Db H-2Db H-2Db

Solvent Solvent (end) Solvent (end)

SSPYSLHYLSSPDSLHYL

AALLNSAVLAALLNSAGL

MAPIDHTTMMAPIDHTAM

Figure 2 | Modelling of mutant peptide–MHCI complexes. Mutant peptideswere modelled into peptide–MHCI structures using existing crystalstructures from the Protein Data Bank as starting models and optimizingthe conformation of the bound mutant peptide using the programFlexPepDock19. H-2Db and H-2Kb MHC haplotypes are shown in green.Modelled peptides are shown in magenta as stick models, with the mutatedresidue highlighted in yellow.

PD-1

TIM

-3

Total CD8 Adpgk+ CD8

1.54 0.114

0.92497.4

0 8.82

26.564.7

7.23 52.3

19.221.3

2.9 79.7

15.91.45

Spleen

Tumour

c

0

20

40

60

80

100

0

10

20

30

PD

-1 M

FI

Per

cen

t

PD

-1+ T

IM-3

+

CD8 T cells

0

2,000

4,000

6,000

8,000 ** *

0.0

0.5

1.0

1.5

2.0

2.5

CD8

Dextr

am

er

0.176

1.33

a b

0.517

0.0934 0.148

0.477

Adpgk Reps1 Dpagt1

0.18 0.09 0.15

1.33 0.52 0.48

0

1

2

3

4

5

6

PepCon

Reps1 Aatf Adpgk Irgq Cpne1 Dpagt1

Adj+Pep

Adj

Per

cen

t d

extr

am

er+

of

CD

8 T

cells

Per

cen

t d

extr

am

er+

of

CD

8 T

cells

** ** **

** **

Prediction + − + − − −

Adpgk Dpagt1 Reps1

Adj+Pep

Adj

Adj+Pep

Adj

Tumour

Spleen

Spleen

Tumour

0102

103

104

105

0102

103

104

105

0 102 103 104 105 0 102 103 104 105 0 102 103 104 105

0102

103

104

105

0102

103

104

105

0 102 103 104 105 0 102 103 104 105

Total Adpgk+

Per

cen

t

PD

-1+ T

IM-3

+

Total Adpgk+ Total Adpgk+

Figure 3 | Immunogenicity of mutated peptides in vivo. a, C57BL/6 micewere immunized with adjuvant (anti-CD40 antibody plus poly(I:C)) alone(Adj) or in combination with a mix of all long mutated peptides from Table 1(50mg each) (Adj1Pep). Med12 peptide was not included because it was notpredicted to bind to MHCI. After two immunizations on day 0 and day 14,spleens were collected at day 21 and CD8 T cells were stained with differentphycoerythrin (PE)-labelled peptide–MHCI dextramers (indicated on x axis)and analysed by flow cytometry. The adjuvant alone (Adj) group is used asthe reference for immunogenicity. Each dot represents individual mouse withbar representing mean. Pooled data from two independent experiments areshown with n $ 6 (Adj) or n $ 9 (Adj1Pep) mice per group. b, Frequency ofAdpgk, Reps1 or Dpagt1–MHCI-specific CD8 T cells among total CD8 T cellsin MC-38 tumour-bearing mice (tumour size ,500 mm3). The graph onthe bottom represents pooled tumour data from ten individual mice. The barsrepresent mean, **P # 0.01 (two-tailed unpaired t-test). c, PD-1 and TIM-3surface expression was measured on Adpgk–H-2Db dextramer1 or total CD8 Tcells from panel b. A representative staining from spleen (top) and tumour(bottom) is shown with the graph on right showing pooled data from five mice.For tumours, PD-1 median intensity of fluorescence (MFI) (gated on PD-11

cells) in the indicated population is also shown. *P # 0.05, **P # 0.01(two-tailed paired t-test).

RESEARCH LETTER

5 7 4 | N A T U R E | V O L 5 1 5 | 2 7 N O V E M B E R 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

Finally, even in this more difficult therapeutic setting, tumour-bearingmice vaccinated with mutated peptides showed remarkable and sus-tainable inhibition of tumour growth compared to untreated control oradjuvant alone groups (Fig. 4j). Thus, simple peptide vaccination withthe predicted neo-epitopes generated sufficient T-cell immunity to rejecta previously established tumour.

The identification of epitopes that drive the immune response in can-cer is essential to the understanding and manipulation of CD8 T-cellimmune responses for clinical benefit. Recent studies in mice and humanshave suggested that tumour-specific mutations probably have a key rolein shaping the anti-tumour response; however, their identification hasremained a challenge2–9. We developed a novel strategy to identify neo-epitopes by combining whole-exome/transcriptome sequencing and massspectrometry analysis, along with a structural prediction algorithm topredict MHCI peptide immunogenicity. This will not only facilitategeneration of novel vaccines but also make it feasible to track tumour-specific T cells, which could be invaluable for prognosis of cancer patientsin this era of cancer immunotherapy.

To be fully useful in the clinic, it will probably be necessary to simplifythe approach further, so as to rely entirely on computational predictions

of peptide binding rather than mass spectrometry. A purely computa-tional approach, including a MHCI-haplotype-specific structural pre-diction, would require only whole-exome/transcriptome sequencing ofa patient’s tumour, which is beginning to be routinely determined.

Online Content Methods, along with any additional Extended Data display itemsandSourceData, are available in the online version of the paper; references uniqueto these sections appear only in the online paper.

Received 27 May; accepted 28 October 2014.

1. Lawrence, M. S. et al. Mutational heterogeneity in cancer and the search for newcancer-associated genes. Nature 499, 214–218 (2013).

2. Matsushita,H. et al.Cancer exome analysis reveals a T-cell-dependent mechanismof cancer immunoediting. Nature 482, 400–404 (2012).

3. DuPage, M., Mazumdar, C., Schmidt, L. M., Cheung, A. F. & Jacks, T. Expression oftumour-specific antigens underlies cancer immunoediting. Nature 482, 405–409(2012).

4. Castle, J. C. et al. Exploiting the mutanome for tumor vaccination. Cancer Res. 72,1081–1091 (2012).

5. van Rooij, N. et al. Tumor exome analysis reveals neoantigen-specific T-cellreactivity in an ipilimumab-responsive melanoma. J. Clin. Oncol. 31, e439–e442(2013).

6. Gros, A. et al. PD-1 identifies the patient-specific CD81 tumor-reactive repertoireinfiltrating human tumors. J. Clin. Invest. 124, 2246–2259 (2014).

0

1

2

3 Control

Adj

Adj+Pep

0

10

20

30

40

50

0

20

40

60

80

0

2,000

4,000

6,000

8,000

10,000

Con

trol

Adj

Adj+Pep

0

25

50

75

100

0 10 20 300

500

1,000

1,500

2,000

0 10 20 30

CD45 CD80

20

40

60

80

100

Control

Adj

Adj+Pep

b

e

f

Days post tumour challenge

Adj Adj+Pep

Tu

mo

ur

vo

lum

e m

m3

Adj Adj+Pep Control

Days after peptide immunization

j i

g

Adj Adj+Pep0.010

0.040

0.160

0.640

2.560

Per

cen

t A

dp

gk–d

extr

am

er+

of

CD

8 T

cells

Adpgk reactivity in blood

h

Per

cent

IFN

-γ+ o

f C

D8 T

cells

Per

cent

TIM

-3+P

D-1

+

Of

CD

8 T

cells

n = 11 n = 2

Tumour Spleen

c

a

Tum

our

vo

lum

e m

m3

Liv

e c

ells

(%

)

Tumour

Per

cent

dextr

am

er+

of

CD

8 T

cells

Spleen

Per

cent

Ad

pg

k–d

extr

am

er+

of

CD

8 T

cells

Control Adj Adj+Pep Adj Adj+Pep

0.05

0.2

0.80

3.20

Non-tumour -bearing

Tumour-bearing

d

* *

*

*

*

0 10 20 30 0 10 20 300 10 20 300

500

1,000

1,500

2,000

n = 4 n = 8

0

2,000

4,000

6,000

8,000

10,000

PD

-1 M

FI

TIM

-3 M

FI

Adpgk Dpagt1 Reps1

Day –21

Immunization with

Adj or Adj+Pep

Day –7

Boost immunization

Adj or Adj+Pep

Day –1

Measure CD8 T-cell

response in blood

Day 0

MC-38 tumour

inoculation

0

500

1,000

1,500

2,000

Day (0)

MC-38 tumour

inoculation

Day (10)

Immunization

Adj or Adj+Pep

Day (17)

TIL analysis

Con

trol

Adj

Adj+Pep

Con

trol

Adj

Adj+Pep

Con

trol

Adj

Adj+Pep

Per

cent

IFN

-γ+ o

f C

D8 T

cells

Con

trol

Adj

Adj+Pep

0

500

1,000

1,500

2,000

0

500

1,000

1,500

2,000

Figure 4 | Vaccination with immunogenicpeptides provides protection and delays tumourgrowth. a–c, C57BL/6 mice were immunized withmutated peptides (Adpgk, Reps1 and Dpagt1,50mg each) and anti-CD40 antibody plus poly(I:C)as adjuvant (Adj1Pep) or adjuvant alone (Adj)followed by inoculation with MC-38 tumour cellsas outlined in a. b, Tumour growth for Adj (n 5 14)or Adj1Pep (n 5 15) groups. Two mice in theAdj1Pep group that showed tumour growth areshown in red. c, Frequency of CD8 T cells in blood(at day 21) stained with Adpgk–H-2Db dextramerand analysed by flow cytometry. The red dots inthe graph correspond to the mice that developedtumours in the Adj1Pep group (in b). Pooled datafrom two independent experiments are shown.d–j, C57BL/6 mice were inoculated with MC-38tumour cells, and 10 days later immunized withmutated peptides (Adpgk, Reps1 and Dpagt1,50mg each) with adjuvant (Adj1Pep), adjuvantalone (Adj) or left untreated (control) as outlinedin d. e, Seven days post immunization, spleen ortumour cells were stained with PE-labelledpeptide–MHCI dextramers for the indicatedpeptide (for tumours) or with Adpgk–H-2Db

dextramer (for spleen) and analysed by flowcytometry. For comparison, non-tumour-bearingmice were also immunized with adjuvant alone(Adj) or adjuvant with peptides (Adj1Pep) andsplenic CD8 T cells were stained with PE-labelledAdpgk–H-2Db dextramer at day 7. f, CD451

or CD81 T cells in tumours shown as percentageof total live cell gate. g, h, Frequency of TIM-31PD-11 cells and mean intensity of fluorescence(MFI) of PD-1 (on PD-11) and TIM-3 (onPD-11TIM-31) among Adpgk–H-2Db

dextramer1 CD8 TILs measured by flowcytometry. i, TILs or splenocytes were stimulatedwith PMA and ionomycin and IFN-c productionin CD8 T cells was determined by intracellularcytokine staining and analysis by flow cytometry.j, Tumour growth in control, Adj or Adj1Pepgroups (n 5 10 each group). e–j, Data arerepresentative of two independent experiments.The bars represent mean. *P # 0.05 (two-tailedunpaired t-test).

LETTER RESEARCH

2 7 N O V E M B E R 2 0 1 4 | V O L 5 1 5 | N A T U R E | 5 7 5

Macmillan Publishers Limited. All rights reserved©2014

7. Tran, E. et al.Cancer immunotherapy based onmutation-specific CD41 T cells in apatient with epithelial cancer. Science 344, 641–645 (2014).

8. Wick, D. A. et al. Surveillance of the tumor mutanome by T cells during progressionfrom primary to recurrent ovarian cancer. Clin. Cancer Res. 20, 1125–1134(2014).

9. Brown, S. D. et al. Neo-antigens predicted by tumor genome meta-analysiscorrelate with increased patient survival. Genome Res. 24, 743–750 (2014).

10. Chen, D. S. & Mellman, I. Oncology meets immunology: the cancer-immunitycycle. Immunity 39, 1–10 (2013).

11. Lundegaard, C. et al. NetMHC-3.0: accurate web accessible predictions of human,mouse and monkey MHC class I affinities for peptides of length 8–11. NucleicAcids Res. 36, W509–W512 (2008).

12. Gujar, S. A., Pan, D. A., Marcato, P., Garant, K. A. & Lee, P. W. Oncolytic virus-initiatedprotective immunity against prostate cancer. Mol. Ther. 19, 797–804 (2011).

13. Fortier, M. H. et al. The MHC class I peptide repertoire is molded by thetranscriptome. J. Exp. Med. 205, 595–610 (2008).

14. Granados, D. P. et al. MHC I-associated peptides preferentially derive fromtranscripts bearing miRNA response elements. Blood 119, e181–e191 (2012).

15. Barbieri, C. E. et al. Exome sequencing identifies recurrent SPOP, FOXA1 andMED12 mutations in prostate cancer. Nature Genet. 44, 685–689 (2012).

16. Bertsch, E.et al.MED12andHMGA2mutations: two independent genetic events inuterine leiomyoma and leiomyosarcoma. Mod. Pathol. 27, 1144–1153 (2014).

17. Goldberg, A. L. Functions of the proteasome: from protein degradation andimmune surveillance to cancer therapy. Biochem. Soc. Trans. 35, 12–17 (2007).

18. Sette, A.et al. Therelationshipbetween class I binding affinity and immunogenicityof potential cytotoxic T cell epitopes. J. Immunol. 153, 5586–5592 (1994).

19. London, N., Raveh, B., Cohen, E., Fathi, G. & Schueler-Furman, O. RosettaFlexPepDock web server–high resolution modeling of peptide-proteininteractions. Nucleic Acids Res. 39, W249–W253 (2011).

20. Rudolph, M. G., Stanfield, R. L. & Wilson, I. A. How TCRs bind MHCs, peptides, andcoreceptors. Annu. Rev. Immunol. 24, 419–466 (2006).

21. Jin, H. T. et al. Cooperation of Tim-3 and PD-1 in CD8 T-cell exhaustion duringchronic viral infection. Proc. Natl Acad. Sci. USA 107, 14733–14738 (2010).

22. Jin, H. T., Ahmed, R. & Okazaki, T. Role of PD-1 in regulating T-cell immunity. Curr.Top. Microbiol. Immunol. 350, 17–37 (2011).

23. Wherry, E. J. T cell exhaustion. Nature Immunol. 12, 492–499 (2011).24. Fourcade, J. et al. Upregulation of Tim-3 and PD-1 expression is associated with

tumor antigen-specific CD81 T cell dysfunction inmelanoma patients. J. Exp. Med.207, 2175–2186 (2010).

25. Fourcade, J. et al. CD81 T cells specific for tumor antigens can be rendereddysfunctional by the tumor microenvironment through upregulationof the inhibitory receptors BTLA and PD-1. Cancer Res. 72, 887–896(2012).

26. Crespo, J., Sun, H., Welling, T. H., Tian, Z. & Zou, W. T cell anergy, exhaustion,senescence, and stemness in the tumor microenvironment. Curr. Opin. Immunol.25, 214–221 (2013).

Supplementary Information is available in the online version of the paper.

Acknowledgements Theauthors thank A. Bruce and J. Murphy for excellent assistancewith artwork.

Author Contributions M.Y. was involved in planning and performing in vivoexperiments, analysingand interpretingdata, and writing the manuscript. S.J. analysedand interpreted whole-exome sequencing and RNA sequencing data, generatedtranslated FASTA database, searched for potential neo-epitopes. Q.T.P. and T.K.C.performed mass spectrometric data analysis and peptide validation. P.L. performedthe structure prediction of the MHCI–peptide complexes. J.T. performed studies withtumour-bearing mice. S.B. performed and analysed FACS studies on tumour lines.C.F. performed and analysed immunizations experiments. Z.M. oversaw RNAsequencing experiments. I.M. assisted with the study design and the preparation of themanuscript. J.F. and T.W. performed MHCI peptide isolation and mass spectrometricanalysis. L.D and J.R.L. oversaw all the work performed, planned experiments,interpreted data and wrote the manuscript.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare competing financial interests:details are available in the online version of the paper. Readers are welcome tocomment on the online version of the paper. Correspondence and requests formaterials should be addressed to L.D. (delamarre.lelia@gene.com) orJ.R.L. (lill.jennie@gene.com).

RESEARCH LETTER

5 7 6 | N A T U R E | V O L 5 1 5 | 2 7 N O V E M B E R 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

METHODSMHCI peptide isolation and mass spectrometric analysis. MHCI peptide pro-filing was conducted for the H-2Kb and H-2Db ‘ligandome’ of two murine cell linesof H-2b-background: TRAMP-C1 (ATCC) and MC-38 (Academisch ZiekenhuisLeiden) as previously described27. Briefly, 1 3 108 cells from each cell line were lysedand MHCI molecules were immunoprecipitated using two different antibodies cou-pled to CNBr-activated Sepharose resin (GE Healthcare Europe, Freiburg, Germany)in order to extract H-2Kb-specific and H-2Db-specific peptides, respectively. Anti-bodies were derived from a local hybridoma culture, H-2Kb-specific antibody fromY3 and H-2Db-specific antibody from B22.249 clones and purified before use. Pep-tides were eluted off antibody-resin by acid treatment27 and purified via ultrafiltra-tion before mass spectrometric analysis.

Peptides were loaded onto an analytical fused-silica micro-capillary column (75mminternal diameter 3 250 mm) packed with 1.7mm C18 (Waters) and separated byreversed-phase chromatography (nanoAcquity UPLC system, Waters, Milford, MA)using a 180 min gradient. The gradient, composed of Solvent A (0.1% formic acidin water) and Solvent B (0.1% formic acid in acetonitrile) went from 10% to 30% Bat a flow rate of 300 nl per minute. The eluted peptides were analysed by data-dependent acquisition (DDA) in an LTQ-Orbitrap Velos hybrid mass spectrometer(Thermo Fisher Scientific, Bremen, Germany) equipped with a nano electrosprayionization source with spray voltage at 2 kV. Mass spectral data was acquired usingmethods comprising of a full scan (survey scan) of high mass accuracy in the Orbi-trap (R 5 30,000 for TOP3, R 5 60,000 for TOP5), followed by MS/MS (profile)scans either in the Orbitrap (R 5 7,500) on the 3 most abundant precursor ions(TOP3) or in the LTQ on the 5 most abundant precursor ions (TOP5) subjected tocollision-induced dissociation (CID). Mass range for selection of doubly chargedprecursors was 400–750 m/z and 800–1,500 m/z for singly charged precursors. Tan-dem mass spectra data acquisition was triggered for ions of charge state 2 in the TOP5methods. There were 2 TOP3 methods, one with MS2 triggered for ion charge state 2and the second method included triggering only singly charged ions. Seven repli-cate injections were performed for each set of samples.

Synthetic peptides corresponding to identify mutant MC-38 and TRAMP-C1antigen peptides were analysed on an LTQ-Orbitrap Elite mass spectrometer (The-rmoFisher, Bremen, Germany) and ionized using an ADVANCE source (Michrom-Bruker, Fremont, CA) at a spray voltage of 1.2 kV. Peptides were injected ontoa 100mm inner diameter capillary column (NanoAcquity UPLC column, 100mm3 100 mm, 1.7mm, BEH130 C18, Waters Corp) and separated by capillary reversephase chromatography on a NanoAcquity UPLC system (Waters Corp). Sampleswere loaded in 0.1% formic acid/2% acetonitrile/98%water and eluted with a gra-dient of 2–90% Buffer B (where Buffer A is 0.1% formic acid/2%acetonitrile/98%waterand Buffer B is 0.1% formic acid/ 2% water/ 98% acetonitrile) at 1.00 ml min21 witha total analysis time of 30 min. Mass spectral data were acquired using a methodconsisting of one full MS scan (375–1,600 m/z) in the Orbitrap at resolution of60,000 M/DM at m/z 400, followed by MS/MS (profile) scans in the Orbitrap atresolution of 15,000 M/DM at m/z 400 of the peptide ions subjected to collisioninduced dissociation (CID).Sequencing. 1mg of total RNA from MC-38 and TRAMP-C1 cancer cell lines wasused to generate RNA-seq libraries using TruSeq RNA sample preparation kit (Illu-mina, CA). Total RNA was purified from cell lines and fragmented to 200–300 bp,with an average length of 260 bp. RNA-seq libraries were multiplexed (two per lane)and sequenced on HiSeq 2000 as per manufacturer’s recommendation (Illumina,CA). 96.1 million (M) RNA fragments were sequenced from MC-38 and 67.5 Mfrom TRAMP-C1. Exome capture was performed using the SureSelectXT mouseexon kit (Agilent, CA). Exome capture libraries were then sequenced on a HiSeq2000 (Illumina, CA) using the HiSeq sequencing Kit (200 cycles). 60 M (2 3 75 bp)exome reads were sequenced from each cell line.Read mapping. Reads were mapped to the mouse genome (NCBI build 37 or mm9)using GSNAP28. Only uniquely mapped reads were retained for further analysis. 76M RNA fragments were uniquely mapped in the MC-38 sample and 54.6 M in theTRAMP-C1 sample. 50.9 M exome fragments in MC-38 and 52 M fragments inTRAMP-C1 were uniquely mapped. To obtain mouse gene models, Refseq mousegenes were mapped to the mm9 genome using GMAP29 and the genomic sequencewas then used for making the gene models. Out of a total of 49.6 M targeted by theexome capture, 49 M fragments were covered by uniquely mapped reads in each cellline. Of these, 97.6% were covered by at least 10 reads in MC-38 (97.1% in TRAMP-C1), and 91.7% were covered by at least 20 reads in MC-38, and 91.2% in TRAMP-C1.Variant calling. Exome-seq based variants were called using GATK30. Variants wereannotated for effects on transcripts using the variant effector predictor tool31. Thus4,285 coding variants were identified in MC-38 and 949 in TRAMP-C1 (Fig. 1b).RNA-seq based variants from the reference mouse genome were called using thesecriteria: variant allele should be supported by at least 2 reads, variant allele frequencywas greater than or equal to 4%, and variant allele was not strand-biased (based onread alignment to the genome, Fisher’s exact test, P , 0.05). A vast majority of the

variants were supported by exactly 2 reads and these accounted for a majority ofvariants below 20% variant allele frequency. As an ad hoc filter, RNA-seq variantswith exactly 2-read support or less than 20% variant allele frequency were filteredout. After overlapping these variants with the exome-based variants, we obtained1,290 coding variants in MC-38 and 67 such variants in TRAMP-C1 (Fig. 1b).Reference database for LC-MS (liquid chromatography-mass spectrometry).For each amino acid variation, a variant whole protein sequence was generated toform a set of putative proteins to serve as a reference database for searching LC-MSspectra. In the absence of haplotype information, multiple variations in the same pro-tein would feature as separate variant proteins in the database. The current data-base was generated without taking into account any haplotype information.Mass spectrometric data analysis. Tandem mass spectral results were submittedfor protein database searching using the Mascot algorithm ver 2.3.02 (MatrixScience,London, UK) against a concatenated target decoy database Uniprot ver 2011_12 ora transcriptome generated FASTA database; comprising of murine proteins andcommon laboratory contaminants such as trypsin. The data was searched with noenzyme specificity, methionine oxidation (115.995 Da), and 20 p.p.m. precursorion mass tolerance. Fragment ion mass tolerance was specified at 0.8 Da or 0.05 Dafor MS/MS data acquired in the LTQ or Orbitrap, respectively. Search results werefiltered using a linear discriminant algorithm (LDA) to an estimated peptide falsediscovery rate (FDR) of 5%. The data was further filtered by peptide length, 8–11amino acids. For the mutated peptide synthetic peptides were generated to validatethe sequences.Modelling of mutant peptide–MHCI complexes. For generation of initial mod-els, peptide–MHC complex structures were chosen from the PDB based on sequencesimilarity between the mutant peptide and peptide in the model structure. For eachmutant peptide model, the following PDB code was used: Reps1, 2ZOL32; Adpgk,1HOC33; Dpagt1, 3P9L34; Cpne1, 1JUF35; Irgq, 1FFN36; Aatf, 1BZ937. The Med12peptide was not modelled due to a lack of a published H-2Kb crystal structure incomplex with 10-mer peptide that could be used as a reasonable starting model.The peptide was then modified to the mutant form using COOT38. These initial modelswere then optimized using the Rosetta FlexPepDock web server19, and the top scor-ing model chosen for display. The top scoring FlexPepDock models for each pep-tide were also inspected, and backbone positioning was found to be similar for thetop ten models generated. Peptide–MHC images were generated using PyMOL(Schrodinger, LLC).Immunization of mice with mutated peptides. Research involving animals com-plied with protocols approved by the Genentech Institutional Animal Care and UseCommittee. Age-matched 6–8 weeks old female C57BL/6 mice (The Jackson Labo-ratory) were injected intraperitoneally with 50mg long peptide each in combinationwith adjuvant (50mg anti-CD40 Ab clone FJK45 plus 100mg poly(I:C) (InvivoGen))in PBS. Mice were immunized on day 0 and day 14 and one week following the lastinjection, either blood or splenocytes were used for detection of Ag-specific CD8T cells. To identify peptide-specific T cells, cells were stained with PE-conjugatedpeptide–MHCI dextramers (MHCI–peptide complex; Immudex, Denmark) for20 min followed by staining with cell surface markers CD3 (catalogue no. 560527),CD4 (catalogue no. 552775), B220 (catalogue no.552772) (BD Biosciences) andCD8 (catalogue no. 100725, Biolegend). Peptide sequences were as follows Reps1:GRVLELFRAAQLANDVVLQIMELCGATR; Adpgk: GIPVHLELASMTNMELMSSIVHQQVFPT; Dpagt1: EAGQSLVISASIIVFNLLELEGDYR; Aatf: SKLLSFMAPIDHTTMSDDARTELFRS; Irgq: KARDETAALLNSAVLGAAPLFVPPAD;Cpne1: DFTGSNGDPSSPYSLHYLSPTGVNEY; Med12: GPQEKQQRVELSSISNFQAVSELLTFE.Tumour-infiltrating lymphocytes. Age-matched 6–8 weeks old female C57BL/6mice were implanted subcutaneously on the right flank with 1 3 105 MC-38 tumourcells. The whole tumour was isolated and digested with collagenase and DNaseto isolate TILs. TILs were stained with peptide–MHCI dextramers (as describedabove) followed by antibodies against CD45 (catalogue no. 561487), CD4 (catalogueno. 553051), CD8 (catalogue no. 553034) (BD Biosciences), Thy1.2 (catalogueno. 140316, BioLegend), PD-1 (catalogue no. 11-9985-82, eBiosciences) and TIM-3 (catalogue no. FAB1529A, R&D Systems). Live/dead stain was used to gate onlive cells.Immunization of tumour-bearing mice. Age-matched 6–8 weeks old femaleC57BL/6 mice animals were inoculated subcutaneously (right hind flank) with 1 3

105 MC-38 cells in a suspension of Hanks’ balanced salt solution (HBSS) and phenolred-free Matrigel (Becton Dickinson Bioscience, San Jose, CA). For prophylacticstudies (in Fig. 4a–c) mice were immunized with adjuvant (50mg anti-CD40 plus100mg poly(I:C) or adjuvant with 50mg Reps1, Adpgk and Dpagt1 peptide each,3 weeks before the tumour inoculation followed by a boost immunization 2 weekslater. Induction of peptide-specific CD8 T cells was measured in blood a day beforeinoculation with tumour cells. For vaccination in tumour bearing mice (Fig. 4d–i),C57BL/6 mice were inoculated with 1 3 105 MC-38 tumour cells and 10 days afterinoculation (only tumour with volume of approximately 100–150 mm3 at day 10

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

were included in the study and were randomly distributed in different groups) micewere injected with adjuvant (anti-CD40 plus poly(I:C)) or adjuvant with 50mg Reps1,Adpgk and Dpagt1 peptides each. Five mice were randomly chosen for analysis byflow cytometry at day 7 post immunization. One mouse with no tumour in Adj1Pepgroup did not yield TILs. For tumour study in Fig. 4j, 30 mg anti-CD40 plus 10 mgpoly(I:C) of adjuvant dose with 50mg Reps1, Adpgk and Dpagt1 peptides each wasused. Blinded tumour measurements and weights were collected twice a week. Ani-mals exhibiting weight loss of more than 15% of their initial body weight wereweighed daily and euthanized if they lose more than 20% of their initial body weight.Animals that exhibited adverse clinical issues were observed more frequently, upto daily depending on the severity, and euthanized if moribund. Mice were eutha-nized if tumour volumes exceeded 3,000 mm3, or after 3 months if tumours did notform. Throughout the entire study, clinical observations of all mice were performedtwice a week.Statistical analysis. Sample sizes were chosen empirically to ensure adequate sta-tistical power and were in the line with accepted standards for the techniques employedin the study. Statistical significance was determined with the un-paired two-tailedStudent’s t-test assuming unequal variance at P # 0.05 level of significance (orindicated in figure legends).

27. Falk, K., Rotzschke, O., Stevanovic, S., Jung, G. & Rammensee, H. G. Allele-specificmotifs revealed by sequencing of self-peptides eluted from MHC molecules.Nature 351, 290–296 (1991).

28. Wu, T. D. & Nacu, S. Fast and SNP-tolerant detection of complex variants andsplicing in short reads. Bioinformatics 26, 873–881 (2010).

29. Wu, T. D.& Watanabe, C.K.GMAP: a genomicmappingand alignmentprogram formRNA and EST sequences. Bioinformatics 21, 1859–1875 (2005).

30. DePristo, M. A. et al. A framework for variation discovery and genotypingusing next-generation DNA sequencing data. Nature Genet. 43, 491–498(2011).

31. McLaren, W. et al. Deriving the consequences of genomic variants with theEnsembl API and SNP Effect Predictor. Bioinformatics 26, 2069–2070 (2010).

32. Butler, N. S. et al. Structural and biological basis of CTL escape in coronavirus-infected mice. J. Immunol. 180, 3926–3937 (2008).

33. Young, A. C., Zhang, W., Sacchettini, J. C. & Nathenson, S. G. The three-dimensionalstructure of H-2Db at 2.4 A resolution: implications for antigen-determinantselection. Cell 76, 39–50 (1994).

34. Denton, A. E. et al. Affinity thresholds for naive CD81 CTL activation by peptidesand engineered influenza A viruses. J. Immunol. 187, 5733–5744 (2011).

35. Ostrov, D. A. et al.How H13histocompatibility peptidesdiffering by a single methylgroup and lacking conventional MHC binding anchor motifs determine self-nonself discrimination. J. Immunol. 168, 283–289 (2002).

36. Wang, B. et al. Peptidic termini play a significant role in TCR recognition.J. Immunol. 169, 3137–3145 (2002).

37. Zhao, R., Loftus, D. J., Appella, E. & Collins, E. J. Structural evidence of T cellxeno-reactivity in the absence of molecular mimicry. J. Exp. Med. 189, 359–370(1999).

38. Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. ActaCrystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004).

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 1 | TRAMP-C1 tumours are less immunogenic thanMC-38 tumours. a, MC-38 or TRAMP-C1 tumours (150–250 mm3 in size)were grown in vivo in C57BL/6 mice. TILs were collected and screened byflow cytometry to quantitate absolute numbers of CD451 immune cells or CD8T cells in each tumour sample. n 5 9 for MC-38 and n 5 5 for TRAMP-C1.

**P # 0.01, ***P # 0.001 by two-tailed unpaired Student’s t-test. b, MHCIsurface expression on MC-38 and TRAMP-C1 cell lines. Cells grown in vitrowere stained with anti-H-2Db (biotinylated) or anti-H-2Kb (PE-conjugated)antibodies or isotype controls and analysed for surface expression by flowcytometry.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 2 | Tandem mass spectra of endogenous mutantMC-38 MHC antigen peptides and their corresponding synthetic peptides.a, AALLNSAVL; b, AQLANDVVL; c, ASM(ox)TNM(ox)ELM(ox), ox,oxidized; d, MAPIDHTTM; e, SSPYSLHYL; f, SIIVFNLL; and g, EEKNTGLI.The peak at 460.9511 m/z corresponds to a co-eluting, singly chargedcontaminant background ion isolated with peptide precursor (EEKNTGLI is an

example of a low-confidence spectral identification, based on comparingthe endogenous and synthetic spectra). Mutation sites are indicated with theamino acid underlined. Spectral differences in endogenous and syntheticpeptide spectra maybe due to co-eluting peptides in the highly complexsample. Although present, not all fragment ions are labelled due to overlappingpeak labels.

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 3 | Variant peptide expression is retained in vivo forthe chosen peptides. Total RNA-seq reads covering the variant position forthe genes corresponding to the 7 MS/MS-identified variant peptides areshown. Both the reference and the alternate alleles are shown for an in vitro andan in vivo sample of the MC-38 line. Expression of the variant allele isobservable both in vitro and in vivo. Alt, alternate allele; Ref, reference allele.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 4 | Tandem mass spectra of endogenous wild-typeMC-38 MHC antigen peptides and their corresponding synthetic peptides.a–c, Wild-type MC-38 peptides AQLPNDVVL (a), ASM(ox)TNRELM(ox),ox, oxidized (b), and SSISNFQAV (c). Spectral differences in endogenous and

synthetic peptide spectra maybe due to co-eluting peptides in the highlycomplex sample. Although present, not all fragment ions are labelled due tooverlapping peak labels.

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 5 | wild-type-counterpart peptides for Adpgk andReps1 are not immunogenic in vivo. a–d, C57BL/6 mice were immunizedwith 100mg wild-type or mutant Adpgk or Reps1 peptides with adjuvant(100mg Poly (I:C) plus 50mg anti-CD40) or adjuvant alone, at day 0 andday 14. At day 21, splenic CD8 T cells from mice immunized withwild-type (GIPVHLELASMTNRELMSSIVHQQVFPT) or mutant(GIPVHLELASMTNMELMSSIVHQQVFPT) Adpgk peptide were stainedwith mut-Adpgk–H-2Db dextramers (a) or WT-Adpgk–H-2Db dextramers(b). Similarly CD8 T cells from mice immunized with wild-type(GRVLELFRAAQLPNDVVLQIMELCGATR) or mutant Reps1

(GRVLELFRAAQLANDVVLQIMELCGATR) peptide were stained withMut-Reps1/H-2Db dextramers (c) or WT-Reps1/H-2Db dextramers (d) andanalysed by flow cytometry. Frequency of indicated peptide–MHCIdextramer1 of CD8 T cells is shown. Mice immunized with adjuvant alone wereused as the reference. Data are from one experiment with n 5 5 each groupwith bars representing means. *P # 0.05 (unpaired two-tailed Student’s t-test).e, CD8 TILs from MC-38 or TRAMP-C1 tumours (300–500 mm3) grownin vivo in C57BL/6 mice were stained with PE-labelled Adpgk–H-2Db

dextramers and analysed using flow cytometry. Frequency of Adpgkdextramer1 among CD8 T cells is shown in two different tumour samples.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 6 | Frequency of peptide-specific CD8 T cells in MC-38 tumours of varying size grown in vivo in C57BL/6 mice. a, b, CD8 T cellsin TILs from tumours of indicated sizes were stained with PE-labelledpooled peptide–MHCI dextramers (Adpgk1Reps11Dpagt1) and analysedby flow cytometry. The graphs are showing percent of peptide–MHCIdextramer1 among CD8 T cells (a) and the frequency of CD8 T cells amongCD451 cells (b) in TILs. Each point represents an individual mouse andthe bars represent means.

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 7 | Frequency of Reps1- and Dpagt1-specific CD8 Tcells in blood after immunization with peptides. Blood CD8 T cells fromC57BL/6 mice were analysed a day before inoculation (at day 21) withMC-38 tumours, after immunization with vaccine as outlined in Fig. 4a. Thefrequencies of peptide-specific CD8 T cells in the adjuvant alone (Adj) or

adjuvant plus peptides (Adj1Pep) groups are shown by staining CD8 T cellswith indicated peptide–MHCI dextramers and analysed using flow cytometry.The two mice that developed tumours in Adj1Pep group (in Fig. 4b) arehighlighted in red filled circles. Pooled data from two experiments areshown with the bars representing means.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 8 | Frequency of Reps1 and Dpagt1-specific CD8 Tcells in spleen after immunization of MC-38 tumour bearing or non-tumour-bearing mice. Tumour bearing or non-tumour-bearing C57BL/6mice were immunized with adjuvant (anti-CD40 plus poly(I:C) alone (Adj) or50mg of Reps1, Dpagt1 and Adpgk peptides with adjuvant (Adj1Pep)) asoutlined in Fig. 4d. a, b, Seven days after immunization splenic CD8 T cells were

analysed for peptide-specific CD8 T cells by staining with Reps1–H2-Db-specific (a) or Dpagt1–H2-Kb-specific (b) dextramers and analysed using flowcytometry. Two groups of non-tumour-bearing mice were also immunized andsplenic CD8 T cells were analysed at day 7 after immunization. Data arerepresentative of two independent experiments and bars represent means.

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 9 | Increased frequency of Adpgk-specific CD8 T cellsand IFN-c-producing CD4 T cells upon immunization of MC-38 tumour-bearing mice. a, Frequency of Adpgk–H-2Db dextramer1 CD8 T cells amongtotal live cells in MC-38 tumours. TILs were analysed after immunization ofMC-38 tumour bearing C57BL/6 mice with adjuvant alone (Adj) or adjuvantwith peptides (Adj1Pep) or no treatment (Control) (as in Fig. 4d). TumourCD8 T cells were stained with PE-labelled Adpgk–H-2Db dextramers and

analysed using flow cytometry. Frequency of Adpgk–H-2Db dextramer1

CD8 T cells of total live cells in tumours is shown. b, Frequency of IFN-cexpressing CD4 T cells in tumours or spleen at day 7-post immunization. As inFig. 4i, TILs or splenocytes were stimulated with PMA and ionomycin andIFNc production among CD4 T cells was determined by intracellular cytokinestaining. Data are representative of two independent experiments with barsrepresenting means.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2014